Deepseek R1 traces from OpenRouter dropping soon in Apollo. Will be the most private way to try full-sized R1 on your phone, if you don’t want to use the DeepSeek app.
LLMS
-
Meta FAIR LLM Projects: OPT, Galactica, and Zetta Development
By
–
You misread.
There had been multiple LLM projects within FAIR for years. Some were open sourced as research prototypes (e.g. OPT175B, Galactica, BlenderBot…).
In mid-2022, FAIR started a large LLM project called Zetta, which was still going in late 2022 when ChatGPT came out.
A -
Perplexity’s Auto Mode for Task-Specific Search
By
–
Perplexity is working on the "Auto" mode, which can automatically switch between Pro search and reasoning depending on the task. An option to "rewrite" a response with a specific model is still available.
-
Qwen Narrows Gap Between Plus and Max
By
–
Qwen has updated its Qwen2.5-Plus model, reducing the performance gap between Plus and Max—the most advanced Qwen model available.
-
Which Generative AI Tool Is Most Essential?
By
–
What's the one generative Al tool you can't live without?
-
New Models Competing with Sonnet for Coding Dominance
By
–
all the new models trying so hard to take the coding throne from sonnet reminds me of this pic.twitter.com/mztO0Q3w5t
— Alex Albert (@alexalbert__) 2 février 2025all the new models trying so hard to take the coding throne from sonnet reminds me of this
-
The shift from traditional search to AI-powered search
By
–
4️⃣ Recherche IA
— Jouhatsu | AI Influence Operator (@Jouhatsu_ai) 2 février 2025
Fini Google Search ! L’ère de la recherche IA est arrivée.
Profitez de la combinaison parfaite entre récupération d’informations en temps réel et réponses intelligentes grâce à Perplexity. pic.twitter.com/5YkGzHqgacRecherche IA Fini Google Search ! L’ère de la recherche IA est arrivée. Profitez de la combinaison parfaite entre récupération d’informations en temps réel et réponses intelligentes grâce à Perplexity.
-
TensorLLM Framework Achieves 250x MHA Weight Compression
By
–
9). TensorLLM Proposes a framework that performs MHA compression through a multi-head tensorisation process and the Tucker decomposition. Achieves a compression rate of up to ∼ 250x in the MHA weights…
-
DeepSeek-R1 Usage Recommendations and Prompting Guide
By
–
6). Usage Recommendation for DeepSeek-R1 This work provides a set of recommendations for how to prompt the DeepSeek-R1 model.
-
O1-like LLMs underthinking patterns and limitations
By
–
4). On the Underthinking of o1-like LLMs Looks more closely at the "thinking" patterns of o1-like LLMs. We have seen a few recent papers pointing out the issues with overthinking.